The FAIR data principles — Findable, Accessible, Interoperable, Reusable — turn ten in 2026. Since Mark Wilkinson and colleagues published the framework in Scientific Data in 2016, FAIR has moved from an aspirational statement of good practice to a hard requirement embedded in funder mandates, journal policies, and institutional research data management infrastructure. UKRI’s open access policy now expects data underpinning publications to be made available in line with FAIR, the US NIH data sharing policy is actively enforced for funded projects, and Horizon Europe applicants must demonstrate FAIR-compliant data management as a condition of award.
Yet a decade in, compliance remains uneven. Many institutions still treat FAIR as a checkbox on a data management plan template rather than a set of concrete technical and governance obligations. As the ten-year anniversary approaches and funders sharpen enforcement, research administrators need a working map from principle to practice — one that goes beyond restating the acronym and instead specifies what each letter actually requires of repositories, metadata schemas, and institutional policy.
This article revisits the original FAIR framework as stewarded by FORCE11 and the GO FAIR initiative, and translates each element into actions that research offices, data stewards, and library services can implement now, ahead of the next REF cycle and continued tightening of funder mandates.
What the FAIR Data Principles Actually Require
Wilkinson et al. (2016) deliberately wrote FAIR as a set of guiding principles rather than a rigid standard, which has allowed broad adoption but also created room for superficial interpretation. FORCE11, the scholarly communication community that convened the original working group, and GO FAIR, the international support and coordination initiative, both continue to publish implementation guidance. For research administrators, the practical translation looks like this:
- Findable — Every dataset needs a globally unique, persistent identifier (a DOI minted through DataCite is the de facto standard for research data) and rich, indexed metadata that describes the dataset independently of the data itself. Institutional repositories must expose this metadata to harvesters and search services, not bury it behind a login wall.
- Accessible — Data (and, critically, its metadata) should be retrievable via a standardised, open communication protocol, with clear authentication and authorisation procedures where restrictions are legitimate. Accessible does not mean “open by default” — it means the access conditions are documented, discoverable, and enforced consistently, even when the data itself is restricted for ethical or commercial reasons.
- Interoperable — Metadata and data should use formal, shared, broadly applicable vocabularies for knowledge representation, and reference other data and metadata using standard identifiers. This is where controlled vocabularies, ontologies, and cross-referencing to identifiers like ORCID (for contributors), ROR (for institutions), and CrossRef (for related publications) matter most.
- Reusable — Data must carry a clear, accessible data usage licence, detailed provenance, and be described with enough domain-relevant metadata that a future researcher — human or machine — can understand and reuse it without contacting the original team.
None of the four elements is optional or substitutable for another. A dataset with a DOI but no licence is findable but not reusable. A dataset described only in free-text notes is accessible but not interoperable. Institutions that treat FAIR as satisfied once a DOI is assigned are addressing roughly one letter out of four.
Persistent Identifiers, Metadata, and Vocabularies: The Infrastructure Layer
The technical backbone of FAIR compliance rests on three infrastructure decisions that research administrators are often best placed to influence, even without deep technical expertise.
First, persistent identifier coverage needs to extend beyond the dataset itself. Contributor identification through ORCID, organisational identification through ROR, and publication linkage through CrossRef and DataCite together create the graph of relationships that makes data genuinely findable and interoperable — not just archived. Institutions that mandate ORCID at the point of data deposit, rather than treating it as optional metadata, see materially better linkage between datasets, grants, and outputs.
Second, metadata schemas need to move beyond generic Dublin Core toward domain-specific standards where they exist — DataCite Metadata Schema as a baseline, supplemented by discipline-specific vocabularies (such as those maintained by biomedical, environmental, or social science data communities). Rich metadata is the single most under-invested element of FAIR compliance: it is unglamorous, resource-intensive to produce well, and rarely rewarded in the same way a publication or citation is.
Third, standard vocabularies and licensing need institutional defaults rather than case-by-case decisions. A repository that offers a menu of Creative Commons or equivalent licences at deposit, with a sensible institutional default and clear guidance on when to deviate, removes the single most common point of friction — researchers who simply skip the licensing step because no default is presented.
From FAIR to CARE: Data Governance Beyond Technical Compliance
FAIR was designed primarily to solve a technical and infrastructural problem: making data machine-actionable and reusable. It says comparatively little about who benefits from that reuse, who consented to it, and who retains authority over data concerning specific communities. This gap is precisely what the CARE Principles for Indigenous Data Governance — Collective Benefit, Authority to Control, Responsibility, and Ethics — were developed to address, and the two frameworks are increasingly discussed together rather than as alternatives.
Institutions building research data governance frameworks in 2026 need to treat FAIR and CARE as complementary rather than competing. FAIR asks “can this data be found, accessed, and reused efficiently?” CARE asks “should it be, on what terms, and who decides?” A research data management policy that only addresses FAIR risks technically excellent infrastructure applied to data — particularly Indigenous, community, or otherwise sensitive data — without adequate governance over consent, benefit-sharing, or ongoing authority. Data governance frameworks that reference both FAIR and CARE principles are becoming standard practice at institutions with significant Indigenous studies, community health, or population genomics portfolios, and reviewers increasingly expect to see both addressed in ethics and data management documentation, not just FAIR.
Building a Research Data Management Plan That Delivers FAIR
The research data management plan is where FAIR principles are supposed to become operational commitments, yet many plans are still written to satisfy a funder template rather than to genuinely guide the research team. A data management plan that actually delivers FAIR outcomes needs to specify, in concrete and checkable terms:
- Which repository will host the data, and whether that repository mints persistent identifiers and supports the metadata schema required for the discipline.
- Who is responsible for metadata creation and quality review before deposit — not left as an afterthought at project close-out.
- Which licence will apply to the data, decided at the planning stage rather than retrofitted at submission.
- What vocabularies or ontologies will be used to describe variables, samples, or methods, particularly where cross-study interoperability is a stated goal.
- How access will be governed for any data subject to ethical, commercial, or CARE-relevant restrictions, including who approves access requests after the project team disbands.
Institutions preparing for REF 2029 and equivalent national assessment exercises have a particular incentive to get this right now: data management practice is increasingly scrutinised as part of research environment statements, and a portfolio of well-governed, genuinely FAIR datasets is a defensible evidence base in a way that a folder of unlinked spreadsheets is not.
What This Means for Research Administrators
For research administrators, EARMA and ARMA members, and institutional research office staff, the ten-year mark for FAIR is a natural prompt to audit rather than assume compliance. Three actions stand out as immediately actionable:
First, audit repository defaults. Check whether your institutional repository mints DOIs automatically, requires a licence selection at deposit, and exposes metadata to standard harvesting protocols. If any of these is missing, that is a findability or reusability gap regardless of how the policy documents read.
Second, build ORCID, ROR, and DataCite/CrossRef linkage into deposit workflows as mandatory fields, not optional extras. This is the lowest-cost, highest-leverage intervention available to most institutions and directly strengthens the Findable and Interoperable pillars.
Third, extend data governance conversations to explicitly include CARE alongside FAIR wherever research involves Indigenous communities, sensitive population data, or community-held knowledge. Reviewers, ethics committees, and increasingly funders are asking for both.
Looking Ahead
As FAIR approaches its tenth anniversary, the framework’s core insight — that data value compounds when it is genuinely findable, accessible, interoperable, and reusable — remains sound. What has changed is the level of scrutiny applied to claims of compliance. Funders, publishers, and institutions themselves are moving from asking “do you have a data management plan?” to asking “does your data actually behave like FAIR data?” For research administrators, closing that gap between policy and practice — with the infrastructure, governance, and plan quality to match — is the work of the next decade, not just the anniversary year.
